Kokoro-82M-v1.0-ONNX

Kokoro-82M-v1.0-ONNX

Turns English text into spoken audio with Kokoro, an 82 million parameter text-to-speech model packaged as ONNX files for JavaScript and Python. For developers who want a small local voice in an app.

Made by onnx-community on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Kokoro is a text-to-speech model with 82 million parameters. You give it English text and it returns audio. This repository holds the ONNX version of the model, so you can run it from JavaScript with transformers.js or from Python with onnxruntime.

The model card calls it a frontier model for its size.

Who it suits: It suits developers who want to add speech output to a web page, a Node app or a Python script without calling a hosted voice service.

Industry | Who uses it | How they use it | Example
Software | App developers | Add a spoken voice to a Node or browser app | Read a notification aloud
Education | Course builders | Turn lesson text into audio files | Make a narrated study sheet
Accessibility | Interface designers | Read page text out loud for users | Voice a long article
Media | Content teams | Produce narration drafts from scripts | Save a short clip as a WAV file

If you need voices in languages other than English, you can skip it, since the card lists English only.

Where to use it: The model runs wherever JavaScript with transformers.js or Python with onnxruntime runs, so Windows, macOS and Linux all work.

• A Node app through the kokoro-js library
• A browser app built on transformers.js
• A Python script that uses onnxruntime

Use case | Who | What they get back
Narration | Content teams | A WAV file with the spoken text
Voice in an app | Developers | Audio generated on the user's side
Voice choice | Designers | A named voice such as af_bella

How to set it up: For JavaScript you need Node and the kokoro-js package from NPM. For Python you need onnxruntime, numpy and the downloaded model files.

• Node with npm, or Python with onnxruntime
• Room for the ONNX file you pick, from full precision to 4-bit versions

1. Install the kokoro-js library with npm.
2. Load the model by its id, and pick a data type.
3. Generate audio from your text and choose a voice.

npm i kokoro-js

How to use it: You load the model once, then call generate with your text and a voice name. The result is an audio object you can save as a WAV file.

import { KokoroTTS } from "kokoro-js";

const model_id = "onnx-community/Kokoro-82M-ONNX";
const tts = await KokoroTTS.from_pretrained(model_id, {
  dtype: "q8", // Options: "fp32", "fp16", "q8", "q4", "q4f16"
});

const text = "Life is like a box of chocolates. You never know what you're gonna get.";
const audio = await tts.generate(text, {
  // Use `tts.list_voices()` to list all available voices
  voice: "af_bella",
});
audio.save("audio.wav");

Tool | What it does
KokoroTTS.from_pretrained | Loads the model with the data type you choose
tts.generate | Turns text into audio with a chosen voice
tts.list_voices | Lists the voices you can pick
audio.save | Writes the audio to a WAV file

Example: You pass a short sentence to the generate call with the voice af_bella and save the result as audio.wav. The file holds that sentence spoken aloud.

For newcomers and experts: If you are new, copy the JavaScript snippet, run it once and play the WAV file it writes. Then change the sentence or the voice name and run it again.

If you know the field, the Python route works one step lower. You turn text into phonemes, map them to token ids, add pad ids at both ends and feed a style vector. The context length is 512 tokens.

Text goes in as English words
It becomes phonemes and token ids
The ONNX model runs with a voice
Audio comes out and is saved

Good to know: The card offers several ONNX builds, such as full precision, fp16, quantized and 4-bit versions, so you can trade size against quality.

• License: apache-2.0
• Cost to run: none named
• Data it reaches: model files you download and the text you pass in
• Not for: other languages, since the card lists English only

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: transformers.js
• Downloads in the last 30 days: 828,320
• Likes: 290
• License: apache-2.0
• Languages: en
• Based on: hexgrad/Kokoro-82M
• Quantized versions: 7
• Fine-tunes: 1
• Tags: transformers.js, onnx, style_text_to_speech_2, text-to-speech

Links

• Model: https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX
• Files: https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download onnx-community/Kokoro-82M-v1.0-ONNX

Then follow the usage in its model card. Nothing in this kit runs by itself.